TxGemma: Open Models To Improve Medicines Development

Google presents TxGemma, a set of open models created to harness the potential of large language models to increase the effectiveness of therapeutic development

[{"selector":"#anim-4d30f375-9c74-4859-a250-60f6cbb295ad","keyframes":{"opacity":[0,1]},"delay":0,"duration":600,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-eca66ef5-5c5c-48f3-b6a9-df8700516d0a","keyframes":{"transform":["translate3d(-115.09434%, 0px, 0)","translate3d(0px, 0px, 0)"]},"delay":0,"duration":600,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-ad804db6-e8fa-4465-b7da-2cd73ac4c239 [data-leaf-element=\"true\"]","keyframes":{"transform":["translate3d(-12.514648210040342%, 0, 0)","translate3d(0%, 0, 0)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}]

TxGemma is trained to understand and predict therapeutic entity attributes throughout the discovery process, from discovering interesting targets to predicting clinical trial outcomes

[{"selector":"#anim-adf34775-cbef-4003-930c-33003d261159","keyframes":{"transform":["scale(1)","scale(1.5)","scale(0.95)","scale(1)"],"offset":[0,0.33,0.66,1]},"delay":0,"duration":1450,"easing":"ease-in-out","fill":"both","iterations":1}] [{"selector":"#anim-e563b26f-3dd6-4950-a18c-5009677ece2d [data-leaf-element=\"true\"]","keyframes":{"transform":["translate3d(12.514648210040342%, 0, 0)","translate3d(0%, 0, 0)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}]

Google unveiled Tx-LLM, a language model trained for a range of therapeutic tasks associated with drug development, last October

[{"selector":"#anim-a2821fe0-a026-4058-abb1-85c2a616161a","keyframes":{"opacity":[0,1]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] [{"selector":"#anim-d02e2402-52da-4a42-9464-7120ad8cd79e [data-leaf-element=\"true\"]","keyframes":{"transform":["translate(0%, 0%) scale(1.5)","translate(0%, 0%) scale(1)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}]

TxGemma has “chat” 9B and 27B. Since they were trained with general instruction tuning data, these models can explain their reasoning, answer complex questions, and participate in multi-turn arguments

[{"selector":"#anim-faf45a6b-cd6a-48a2-bef4-53c71d3072d7","keyframes":{"transform":["scale(1)","scale(1.5)","scale(0.95)","scale(1)"],"offset":[0,0.33,0.66,1]},"delay":0,"duration":1450,"easing":"ease-in-out","fill":"both","iterations":1}] [{"selector":"#anim-96cde725-540f-4632-9290-98b63e186fe2 [data-leaf-element=\"true\"]","keyframes":{"transform":["translate(0%, 0%) scale(1.5)","translate(0%, 0%) scale(1)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"forwards"}]

TxGemma may be included into agentic systems to address more challenging research issues than single-step forecasts

[{"selector":"#anim-a7aeaa0f-ec25-45f0-8f00-2abe0a0a2ce1","keyframes":{"transform":["translate3d(-115.09434%, 0px, 0)","translate3d(0px, 0px, 0)"]},"delay":0,"duration":1000,"easing":"cubic-bezier(.2, 0, .8, 1)","fill":"both"}] [{"selector":"#anim-3f9d357e-7032-464c-b872-84e5d0215df5","keyframes":{"transform":["rotateZ(-180deg)","rotateZ(0deg)"]},"delay":0,"duration":1000,"easing":"cubic-bezier(.2, 0, .5, 1)","fill":"forwards"}] [{"selector":"#anim-72c5b331-e576-441d-99c5-0424ab4e59b0 [data-leaf-element=\"true\"]","keyframes":{"transform":["translate3d(-12.514648210040342%, 0, 0) translate(-25%, 0%) scale(1.5)","translate3d(0%, 0, 0) translate(0%, 0%) scale(1)"]},"delay":0,"duration":2000,"fill":"forwards"}]

Standard language models struggle with multi-step thinking and current external knowledge. Google designed Agentic-Tx, a Gemini 2.0 Pro-powered therapeutics agentic system

[{"selector":"#anim-3e609c2a-12c1-4d42-ba90-0aee4095a268","keyframes":{"opacity":[0,1]},"delay":0,"duration":600,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-ab2ddcf8-25cd-4da4-83b1-3e789721b020","keyframes":{"transform":["translate3d(0px, -120.60338%, 0)","translate3d(0px, 0px, 0)"]},"delay":0,"duration":600,"easing":"cubic-bezier(0.2, 0.6, 0.0, 1)","fill":"both"}] [{"selector":"#anim-f622218c-dded-47f0-a8f3-4720a53b58e2 [data-leaf-element=\"true\"]","keyframes":{"transform":["translate3d(-12.514648210040342%, 0, 0)","translate3d(0%, 0, 0)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}]

TxGemma is currently available on Hugging Face and Vertex AI Model Garden

[{"selector":"#anim-89747b42-c75f-4700-9ccf-2756a538d72d [data-leaf-element=\"true\"]","keyframes":{"transform":["translate3d(-12.514648210040342%, 0, 0)","translate3d(0%, 0, 0)"]},"delay":0,"duration":2000,"easing":"cubic-bezier(.3,0,.55,1)","fill":"both"}] [{"selector":"#anim-8169dfa1-998d-453f-8ba7-16075ae632bc","keyframes":{"transform":["translate3d(119.12751%, 0px, 0)","translate3d(0px, 0px, 0)"]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] [{"selector":"#anim-7bcc6021-82a2-4eaa-9630-78af1613e622","keyframes":{"opacity":[0,1]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"both"}] [{"selector":"#anim-8d432448-481d-45ff-8768-9ad73bf1ca31","keyframes":{"transform":["scale(0.15)","scale(1)"]},"delay":0,"duration":600,"easing":"cubic-bezier(0.4, 0.4, 0.0, 1)","fill":"forwards"}]