David R. Liu ✓

@davidrliu · Massachusetts, USA · joined 28 Nov 2008

Professor at Harvard and the Broad Institute. Our lab integrates chemistry & evolution to illuminate biology & enable new therapeutics. IG/Threads: @davidrliu1

80 125Followers

69Following

3 581Posts total

340.2KViews on collected posts

โพสต์ล่าสุด

David R. Liu ✓@davidrliu12 Aug

@davidrliu @NatureBiotech Congratulations! Exciting work. Your lab is putting out so much good stuff, kudos! 441 views · 3 likes · 0 reposts · 0 replies Open on X →

David R. Liu ✓@davidrliu12 Aug

This work was led by Alvin Hsu, @PeterJ_Chen, and Angus Li, with key contributions from Colin Hemez, @danielxingao, Markus Terrey, Charlie Nelson, Vijay Selvam, @Ana_Cristian_, Amber McElroy, Benjamin Steinbeck, Gandhar Mahadeshwar, Smriti Pandey, @zacharybarsdale, @paulchenz, 1.4K views · 13 likes · 0 reposts · 0 replies Open on X →

David R. Liu ✓@davidrliu12 Aug

As a resource to the community, we make OptiPrime freely available for non-commercial use at https://t.co/XKquv1LE4J. The web interface supports automated lookup of genetic variants documented in ClinVar and dbSNP, as well as manual entry of HGVS notation, genomic coordinates, or 1.5K views · 6 likes · 0 reposts · 1 replies Open on X →

David R. Liu ✓@davidrliu12 Aug

To test this rapidly developed PE strategy in vivo, we used the v1em dual-AAV architecture previously developed in our group by @JessieRoseDavis and @samagyabanskota. Intracerebroventricular (ICV) injection of AAV9s encoding the lead PE strategy into neonatal Kif1a “legdragger” h 670 views · 3 likes · 0 reposts · 1 replies Open on X →

David R. Liu ✓@davidrliu12 Aug

As a proof-of-concept, we used OptiPrime to design just 8 pegRNAs to correct Kif1a “legdragger” (p.L181F), a mutation that causes spastic paraplegia in homozygous mice similar to symptoms of patients with pathogenic mutations in KIF1A. We performed experiments directly in mouse h 1.3K views · 4 likes · 1 reposts · 1 replies Open on X →

David R. Liu ✓@davidrliu12 Aug

Similarly, we found OptiPrime-nominated editing strategies for the correction of COL7A1 p.R185X, which causes recessive dystrophic epidermolysis bullosa (RDEB), and for installation of the IL2RB p.H134D Y135F variant, which enables IL-2 receptor stimulation by an orthogonal https 704 views · 3 likes · 0 reposts · 1 replies Open on X →

David R. Liu ✓@davidrliu12 Aug

A team lead by @kiranmusunuru and @AhrensNicklas that included our lab recently reported the treatment of baby K.J. Muldoon’s CPS1 deficiency with a customized base editor—the first personalized gene editing therapy. Inspired by this example, we investigated whether OptiPrime htt 1.2K views · 8 likes · 1 reposts · 1 replies Open on X →

David R. Liu ✓@davidrliu12 Aug

Next, we sought to evaluate OptiPrime’s ability to streamline the identification of effective editing strategies in a variety of therapeutically relevant use cases. OptiPrime-nominated pegRNA designs for the correction of CFTR p.F508del, the most common cause of cystic fibrosis, 1.3K views · 6 likes · 1 reposts · 1 replies Open on X →

David R. Liu ✓@davidrliu12 Aug

We reasoned that we could use OptiPrime’s understanding of effective rates to predict the efficiencies of prime editing variants not present in OptiPrime’s training data. We tested OptiPrime predictions for informing the design of twin prime editing (twinPE) strategies, which htt 764 views · 5 likes · 0 reposts · 1 replies Open on X →

David R. Liu ✓@davidrliu12 Aug

We also found that including OptiPrime predictions in models for PE3, which incorporates an additional nicking guide to stimulate repair of the non-edited strand, substantially improved performance compared to models that utilized only predictive scores for Cas9 nuclease activity 819 views · 5 likes · 0 reposts · 1 replies Open on X →

David R. Liu ✓@davidrliu12 Aug

When designing OptiPrime, we hypothesized that a model architecture in which the PE mechanism is directly incorporated into the mathematical structure of the model would improve performance and generalization. OptiPrime constructs a system of differential equations that describes 956 views · 6 likes · 1 reposts · 1 replies Open on X →

David R. Liu ✓@davidrliu12 Aug

In ablation studies, we found that OptiPrime’s model architecture enabled superior performance compared to model variants that did not correspond as closely to the hypothesized prime editing mechanism. When trained and evaluated on the same data, OptiPrime also predicted https:// 885 views · 5 likes · 0 reposts · 1 replies Open on X →

David R. Liu ✓@davidrliu12 Aug

We used the resulting data to probe how different mismatches respond to MMR by examining how much inhibiting MMR (via PE4) improves PE efficiency over the PE2 baseline. We found that longer edits are generally better at evading MMR. We also determined the MMR propensity for https 1.2K views · 7 likes · 0 reposts · 1 replies Open on X →

David R. Liu ✓@davidrliu12 Aug

In previous work led by @PeterJ_Chen, we discovered that DNA mismatch repair (MMR) opposes PE efficiency. We created a training dataset to better understand the sequence determinants of MMR, enabling the design of prime editing strategies that create longer, MMR-evasive https://t 2.5K views · 20 likes · 2 reposts · 1 replies Open on X →

David R. Liu ✓@davidrliu12 Aug

Today in @NatureBiotech we report OptiPrime, a mechanism-based machine learning model for predicting prime editing (PE) efficiencies and suggesting pegRNAs to achieve a user-specified edit. OptiPrime incorporates biochemical knowledge about PE into its mathematical structure, 73.6K views · 429 likes · 112 reposts · 5 replies Open on X →

David R. Liu ✓@davidrliu12 Aug

Previous efforts to apply machine learning to the prediction of prime editing (PE) outcomes have relied on training regression models on top of embeddings from sequence-based models, whose structures are inspired by natural language processing. This approach typically does not ht 3.3K views · 20 likes · 0 reposts · 1 replies Open on X →

David R. Liu ✓@davidrliu3 Aug

Today in @NatureBiotech we report the use of phage-assisted evolution (PACE) to reprogram botulinum neurotoxin (BoNT) proteases to trigger cancer cell death. We evolved proteases to cleave procaspase-1 or gasdermin D, inducing cell death and slowing tumor growth in a highly https 35K views · 358 likes · 81 reposts · 7 replies Open on X →

David R. Liu ✓@davidrliu1 Aug

Amazing new work by @JShendure and coworkers uses a clever application of prime editing called DNA Typewriter to trace the lineage of a developing mouse from zygote to middle of day 14 at a single-cell level. See Jay’s beautiful tweetorial: 47.8K views · 292 likes · 47 reposts · 0 replies Open on X →

David R. Liu ✓@davidrliu22 Jul

@davidrliu @Nature I'm not understanding how proteinmpnn resequencing leads to improved 'stabiliy' if your masking any residues nearby the mutations - proteinmpnn residue probability distribution does not seem to care about changes in residues at a distance. Is it just by creati 691 views · 1 likes · 0 reposts · 0 replies Open on X →

David R. Liu ✓@davidrliu22 Jul

This work was led by @NicholasKrasnow, with key contributions from Joy Xu, @emilyzhang326, Gandhar Mahadeshwar, @Allentaoyz, Julia McCreary, Colin Hemez, Logan Brown, and Wei Jiang. Congratulations to the authors! The paper is available on the @Nature website: 2.1K views · 15 likes · 1 reposts · 0 replies Open on X →

David R. Liu ✓@davidrliu22 Jul

Finally, we applied these findings to evolve proteases that specifically cleave the disease-relevant protein ataxin-2. Redesign-evolved proteases had higher expression, stability, and ataxin-2 cleavage rates than those evolved from wild-type. Importantly, the best https://t.co/Jn 2.7K views · 20 likes · 2 reposts · 2 replies Open on X →

David R. Liu ✓@davidrliu22 Jul

Each redesigned protein background is compatible with most mutation sets evolved in each of the alternative starting points, while the WT protein background rarely functions with redesign-evolved mutation sets. Notably, the solutions functional in redesign backgrounds include htt 1.2K views · 7 likes · 1 reposts · 1 replies Open on X →

David R. Liu ✓@davidrliu22 Jul

We next explored how alternative redesigns evolve. We carried out the same panel of PACE selections with a hyperstable but activity-impaired MPNN redesign D4, and a redesign from physics-based modelling (non-AI), PROSS1. These redesigns, like D3, raised fitness and expanded the h 1.2K views · 7 likes · 0 reposts · 1 replies Open on X →

David R. Liu ✓@davidrliu22 Jul

We compared the progression of genotype distributions for each starting point. WT protease-initiated lagoons rapidly converged on multi-mutant genotypes, whereas single mutants dominated the redesign-initiated lagoons at early timepoints. The redesign-initiated lagoons converged 1.3K views · 9 likes · 1 reposts · 1 replies Open on X →

David R. Liu ✓@davidrliu22 Jul

In addition to end-state activity, we also compared the speed at which each starting point adapts to the selection. The redesigned starting point evolves faster than its natural counterpart when redesign confers function to local mutations that are rapidly sampled. 9/14 https://t 1.4K views · 7 likes · 1 reposts · 1 replies Open on X →

David R. Liu ✓@davidrliu22 Jul

Across all three selections, the stabilized, activity-enhanced redesign D3 evolved higher activity to cleave the target substrate than the wild-type counterpart. The two starting points evolved distinct genotypes. Strikingly, we found that the redesigned enzyme functions with htt 1.5K views · 10 likes · 0 reposts · 1 replies Open on X →

David R. Liu ✓@davidrliu22 Jul

In a second workflow, we investigated the performance of redesigned enzymes as evolutionary starting points. We hypothesized that the greater stability of redesigned proteases may enhance evolvability on a new substrate by offering compatibility with mutations that improve https: 1.6K views · 8 likes · 0 reposts · 1 replies Open on X →

David R. Liu ✓@davidrliu22 Jul

In a related study, we applied this principle to develop a new suite of prime editors, PE8. ProteinMPNN redesign of PACE-evolved reverse transcriptases improves intracellular levels and gene editing efficiency of prime editor proteins, especially in transient, therapeutically htt 1.8K views · 13 likes · 0 reposts · 1 replies Open on X →

David R. Liu ✓@davidrliu22 Jul

ProteinMPNN redesign of a BoNT/E protease previously evolved in PACE to cleave PTEN substantially elevated expression and potency while retaining the evolved specificity. Redesigned proteases expressed at >24-fold higher levels than either the evolved enzyme or the wild-type http 1.9K views · 6 likes · 1 reposts · 1 replies Open on X →

David R. Liu ✓@davidrliu22 Jul

We use these redesigned BoNT proteases to develop two workflows to enhance protein evolution. In the first, we redesign reprogrammed enzymes after laboratory evolution to restore expression while preserving evolved function. 4/14 https://t.co/jAi4PRLm0D 2K views · 7 likes · 0 reposts · 1 replies Open on X →

เทียบกับบัญชีขนาดเดียวกัน

17 โพสต์จาก 90 วันที่ผ่านมา เทียบกับช่วง 10K–100K ผู้ติดตาม ต่ำกว่าบัญชีรุ่นเดียวกันทั้งการเข้าถึงและการมีส่วนร่วม.

ยอดดูมัธยฐาน1 216บัญชีนี้924ค่ามัธยฐานของ 10K–100K

การเข้าถึง, %1.52%บัญชีนี้3.62%ค่ามัธยฐานของ 10K–100K

การมีส่วนร่วม, %0.69%บัญชีนี้1.52%ค่ามัธยฐานของ 10K–100K

ตัวชี้วัด	บัญชีนี้	ค่ามัธยฐานของ 10K–100K	อัตราส่วน
ยอดดูมัธยฐานต่อโพสต์	1 216	924	1.32×
การเข้าถึง (ยอดดู ÷ ผู้ติดตาม)	1.52%	3.62%	0.42×
อัตราการมีส่วนร่วม	0.69%	1.52%	0.45×

บัญชีอื่นในช่วงนี้ → เปรียบเทียบกับบัญชีอื่น → ค่าอ้างอิงเหล่านี้คำนวณอย่างไร →

Growth & engagement

How the posts we collected actually performed: views and reaction rate post by post, what the audience did with them, and where the follower count goes.

Views per post

2.5K12 Aug

1.2K

885

956

819

764

1.3K

1.2K

704

1.3K

670

1.5K

1.4K

441

Last 14 collected posts, oldest on the left. The scale is logarithmic: one post can outrun the rest a hundred times over.

Engagement rate per post

0.92%12 Aug

0.69%

0.68%

0.84%

0.73%

0.79%

0.62%

0.82%

0.57%

0.45%

0.60%

0.46%

0.92%

0.68%

Reactions — likes, reposts, replies and quotes — divided by views. Median for 10K–100K accounts is 1.52%.

What the audience does

Likes64.5%1 303 in total

Reposts12.5%253 in total

Replies1.8%36 in total

Quotes0.8%17 in total

Bookmarks20.3%411 in total

Share of every reaction we collected for this account. Replies mean argument, reposts mean endorsement, bookmarks mean the post was worth keeping.

The follower curve appears once this account has two daily snapshots — we take one a day, and this one is on its first.

โพสต์ล่าสุด

เทียบกับบัญชีขนาดเดียวกัน

Growth & engagement

Views per post

Engagement rate per post

What the audience does

บัญชีที่คล้ายกัน